Google lets Street View recognize house numbers with 90 percent accuracy

Developers at Google have said they have improved an algorithm for recognizing house numbers in Street View images. Thanks in part to the help of input via its reCaptcha service, the algorithm would recognize up to 90 percent of difficult-to-read house numbers.

For house number recognition, Google uses OCR algorithms developed to analyze photos from StreetView, part of Google Maps, and reCaptcha, an anti-spam security system. The latter partly uses photos of house numbers to display captchas. ReCaptchas are shown more than a hundred million times every day.

Now, thanks to the use of neural network technology and improved algorithms, Google says it is able to recognize up to 90 percent of difficult-to-read multi-digit house numbers, a separate dataset that it has examined, from images shot for Street View. By also presenting difficult-to-read house numbers to internet users via reCaptcha, the recognition would even be increased to 99.8 percent.

The improved OCR algorithms could also solve up to 99 percent of existing text captchas, according to Google. Nevertheless, the study is said to have helped improve its reCaptcha service, in part by showing fewer text captchas to Internet users.